Automatic annotation extension and classification of documents using a probabilistic graphical model

Abdessalem Bouzaieni (1) , Sabine Barrat (2) , Salvatore Tabbone (1)

1 QGAR - Querying Graphics through Analysis and Recognition
2 LIFAT - Laboratoire d'Informatique Fondamentale et Appliquée de Tours

With the fast growth of document images, the domain of document annotations has become a research area of great interest. Annotations allow to describe the semantic content of documents, and facilitate the use and research task for the user. However, for a huge number of documents it is a tedious task to annote each document manually. A solution is to annote a small part of the documents and to extend this annotation automatically to the whole dataset. In this paper, we propose a model for annotation extension and for documents classification using a probabilistic graphical model. In this model, we combine visual and textual characteristics and we show that the integration of the user feedback improves significantly the results.

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]

Abdessalem Bouzaieni, Sabine Barrat, Salvatore Tabbone. Automatic annotation extension and classification of documents using a probabilistic graphical model. 13th International Conference on Document Analysis and Recognition (ICDAR 2014), Aug 2015, Nancy, France. ⟨10.1109/ICDAR.2015.7333775⟩. ⟨hal-01254933⟩